Hearable Computing for Fitness and Health

standard headphones, for example for tracking fitness activities such as running. Following the pandemic they have also seen common use across virtual meetings.
However, although hearables already exist on the market in some form, and are used across a small range of extended applications beyond listening to music, the majority of available devices' functions are generally restricted to means of transmission of audio and speech. Their ability to detect physiology, especially under motion is also still being researched, with significant challenges discovered in considering head and face macro and micro movements. Additionally, they are not generally possibly to use as standalone devices and are usually dependent on smartphones for further computation and communication.
In this project we will lay the foundation to allow hearable devices to become robust, trustable means of detecting and tracking human behaviour, health and fitness. Specifically, we will look at how to allow hearable devices to become stand-alone devices for health and fitness sensing through the investigation of on device machine learning and continuous learning, as well as novel sensor instrumentation.
In order to enable hearables to be used effectively for tracking health and fitness across periods of time, on-device continual learning methods will be developed, allowing systems to operate and improve independently over significant timespans without the necessity of another device such as a smartphone to perform computation. Since hearables are intrinsically relatively small devices, and hence contain space-constrained electronics, this will require investigation of more efficient continual learning techniques which are able to function on very limited hardware, while still maintaining accuracy for critical health data. Class-incremental continual learning will be explored as well, allowing hearable devices to expand their abilities over time in a way that is personalisable for a user, for example...